Our approach is to use real-time input to dynamically generate temporal probabilistic networks, using knowledge-based model construction techniques. Our model generation algorithm will control the size, focus, and fidelity of the models based on available computational resources and salience of model features given the current situation. We exploit compilation techniques to maintain the model at a reasonable size while accumulating information from diverse sensors over time.
Situation assessment from probabilistic models can be employed for traffic management, emergency response, near-accident detection for intersection safety analysis, and intelligent traffic signals, among other applications. Decision-theoretic models can also be applied to the intelligent control of IVHS vehicles, as in the ongoing BAT project. Such control modules can be used to provide high-fidelity models of human drivers for full traffic simulation, with significant applications for highway operations and design.